Intravoxel incoherent motion diffusion-weighted MR imaging for differentiation of benign and malignant musculoskeletal tumours at 3 T

Abstract

Objective:

To evaluate the intravoxel incoherent motion (IVIM) diffusion-weighted (DW) MRI for differentiating between benign and malignant musculoskeletal tumours at 3 T.

Methods:

65 patients with treatment-naïve musculoskeletal tumours (47 malignant and 23 benign lesions) who underwent 3 T MRI including IVIM DW imaging were included. IVIM-derived parameters included pure diffusion coefficient (D), perfusion related incoherent microcirculation (D*, pseudodiffusion coefficient), and perfusion fraction (f). IVIM parameters and mono-exponential apparent diffusion coefficient (ADC) were retrospectively measured by two independent musculoskeletal radiologists.

Results:

D and ADC values of malignant tumours (923 ± 360, 965 ± 353 µm² s^–1, respectively) were significantly lower than those of benign tumours (1668 ± 546, 1689 ± 526 µm² s^–1) (p < 0.001). F values of malignant tumours (9.6%) were significantly higher than those of benign tumours (7.2%) (p = 0.021), whereas D* values showed no significant difference (p > 0.05). The area under the receiver operating characteristic (ROC) curve of D, ADC and f were 0.874, 0.880 and 0.671, respectively. Using cut-off values of D and ADC of 1200 µm² s^–1, the sensitivity, specificity and accuracy were 92, 83, 89%, 92, 87 and 90%, respectively.

Conclusion:

D and ADC may be more accurate and reliable for differentiation of malignant from benign musculoskeletal tumours than f and D* at 3 T IVIM DW imaging.

Advances in knowledge:

Among IVIM-derived parameters, D is more accurate and reliable in differentiating malignant from benign musculoskeletal tumours than f and D* at 3.0T IVIM DW imaging. There was no significant difference in the diagnostic performance of D and ADC.

INTRODUCTION

It is not uncommon that we encounter the problematic cases in differentiation between benign and malignant musculoskeletal tumours in daily practice. There have been inconsistent reports using diffusion-weighted (DW) MRI for differentiating malignant from benign musculoskeletal tumours at 1.5 T.^1–6 However, apparent diffusion coefficient (ADC) can be influenced not only by molecular diffusion but also through microcirculation; therefore, ADC might have a limitation in charactering the lesions.^7,8 Recently, the intravoxel incoherent motion (IVIM) DW imaging has drawn attention. Using the IVIM model and sufficiently low multiple b-values (<200 s mm^–2), microcirculation or perfusion related effects can be separated from pure tissue diffusion (D), and then perfusion characteristics (pseudodiffusion coefficient, D*) and their volume fraction (perfusion fraction, f) can be derived.^8–11 Several previous studies^7,11–17 have introduced the feasibility or diagnostic capacity of IVIM DW imaging in malignant tumours of pancreas, liver, kidney, prostate, salivary gland and breast. In the previous report⁵ using early IVIM DW imaging with five b-values (0, 176, 351, 526 and 701 s mm^–2) at 1.5T was investigated in malignant and benign soft tissue tumours and suggested the significance of true diffusion coefficient. After a decade, according to the report by Koh et al,¹⁸ at least 6–8 b-values were recommended in total for practically optimal distribution: four or more b-values being recommended within the perfusion-sensitive range (<100–200 s mm^–2) and fewer high b-values due to its stability and reproducibility. Thus, perfusion-sensitive information might be insufficient using two b-values less than 200 s mm^–2. 19 There have been few reports regarding IVIM DW imaging with sufficient b-values for differentiating malignant from benign musculoskeletal tumours at 3 T MRI.

The purpose of our study was to evaluate the IVIM DW MRI for differentiating between benign and malignant musculoskeletal tumour at 3 T.

METHODS AND MATERIALS

Study population

Our institutional review board approved this retrospective study and waived the patient informed consent. As a study coordinator, one musculoskeletal radiologist (J.Y.J.) with 6 years of musculoskeletal image interpretation experience collected the MRI and reviewed clinical history. Between August 2013 and March 2014, a total of 249 patients underwent 3 T MRI with IVIM-DW imaging for suspected musculoskeletal tumours in our institution (Figure 1). MR images of 184 patients were excluded according to the following exclusion criteria: (1) patients treated by chemotherapy or radiation therapy (n = 77); (2) patients with musculoskeletal tumours less than 1.0 cm in diameter (n = 50); (3) patients with unsatisfactory image quality or artefacts (n = 17); (4) patients with presumed benign tumours without more than 6 months follow-up (n = 13); (5) patients with presumed malignancy but who did not have pathological confirmation (n = 2); (6) patients with lipoma (n = 18) or vertebral haemangioma (n = 7), which are known as having poor outcome of DW imaging due to a fat portion.^{1, 11, 20} If presumed benign tumours without pathological confirmation showed no change on 6 months follow-up imaging and no occurrence of symptom, it was considered benign. Finally, 65 patients (mean age, 48.4 years; age range, 12–75 years) with 70 musculoskeletal tumours were included in our study. They were 35 males (mean age, 47.5 years; age range, 12–75 years) and 30 females (mean age, 49.4 years; age range, 19–74 years). All tumours had pathological confirmation except one case of fibrous dysplasia in femur.

Figure 1.

Flowchart showing the selection of the study population. DWI, diffusion-weighted imaging; Dx, diagnosis; F/U, follow up; IVIM, intravoxel incoherent motion; MSK, musculoskeletal.

MRI protocols

All patients underwent MR examination with 3 T MR imagers (MAGNETOM Verio, Siemens Healthcare, Erlangen, Germany). Standard MRI was obtained following our clinical protocol with phased-array coils or an eight-channel extremity coils depending on the anatomic regions: longitudinal fat-suppressed T₂ weighted turbo spin-echo (TSE) sequence, axial T₁ weighted TSE sequence, axial T₂ weighted TSE sequences with and without fat-suppression and longitudinal and axial fat-suppressed contrast-enhanced T₁ weighted TSE sequences (Table 1).

Table 1.

MRI parameters

Parameters	Standard sequences	DWI (single shot)
Field of view	80–220 mm	80–220 mm
Matrix size	512 × 256	64 × 45–120 × 128
TR (ms)/	T1 weighted images: 680 - 870/11–21	5000 - 8700/71–85
TE (ms)	T2 weighted images: 4000 - 5600/63–83
Fat suppression	CHESS pulse	CHESS pulse
Section thickness	2–5 mm	2–5 mm
Intersection gap	No	No
Turbo factor or	T1weighted image: 3	56
EPI factor	T2 weighted image: 13
Number of excitation	1	3–5

CHESS, chemical shift selective; DWI, diffusion-weighted imaging; EPI, echo-planar imaging; TR/TE, relaxation time/echo time.

DW images were obtained prior to the contrast material administration by using a single-shot spin-echo echo-planar imaging pulse sequence prototype. The parallel imaging technique (generalized autocalibrating partially parallel acquisitions) with a two-fold acceleration factor was used to shorten echo train length. Encoding was performed in three orthogonal directions. A series of nine b-values (0, 25, 50, 75, 100, 200, 300, 500 and 800 s mm^–2) was applied. The acquired DW imaging data was post-processed to obtain ADC and the IVIM-derived parameters including D, D* and f by using prototype software provided by the manufacturer (Siemens Healthcare, Erlangen, Germany). IVIM-derived parameters were calculated on a voxel-by-voxel basis by a fully bi-exponential non-linear fit according to the following equation proposed by Le Bihan et al:⁸

$${\displaystyle \frac{S(b)}{S_0}=(1-f)\times e{}^{-bD}+f\times e{}^{-b{D}^{\ast }}}$$

Each variable of this equation was defined as follows: S, mean signal intensity according to b-value; f, perfusion fraction or proportion of protons linked to microcirculation; D, diffusion coefficient; D*, pseudodiffusion.

MRI analysis

MR images were retrospectively assessed by two independent musculoskeletal radiologists (W,H,J., H.K.L. with 17 and 5 years of experience of musculoskeletal radiology, respectively). IVIM-derived parameters and ADC were retrospectively measured based on standard MRI to confirm the location and size of the lesion. Region of interest (ROI) was traced manually within the solid portion of the tumour based on standard MRI images, which showed hyperintense signal on high b-value DW images to encompass as much of the lesion as possible excluding the area of haemorrhage or necrosis on the representative section of ADC map. Most peripheral portion, artefacts, image distortions, and a partial volume effect were avoided. And then, the ROI was copied and pasted to the corresponding images of IVIM parameters. The average values of the two ROIs from two readers were respectively calculated to represent the lesion in IVIM-derived parameters and ADC maps. The bi-exponential signal decay curves were obtained using Matlab (MathWorks, Natick, MA).

Statistical analysis

Shapiro–Wilk test was performed for normality. ADC, D and f were non-parametric data, and Mann–Whitney U test was performed for comparison between benign and malignant musculoskeletal tumours, while two sample t-test were performed for comparison of D* that was parametric data. The diagnostic performance of the parameters was evaluated by using receiver operating characteristic (ROC) analysis. The areas under the ROC curve (AUC) were compared for significant difference between ADC and IVIM parameters. Interobserver agreement for the measurement was evaluated by the intraclass correlation coefficient (ICC) using a two-way mixed model with absolute agreement.²¹ ICC of less than 0.40 was indicative of poor agreement; 0.40–0.75, fair to good agreement; and more than 0.75, excellent agreement.²² P-value less than 0.05 was considered statistically significant. The statistical analyses were performed using SPSS v. 20.0 (SPSS Inc., Chicago, IL) and MedCalc v. 14.8.1.0 (MedCalc, Mariakerke, Belgium).

RESULTS

Among 70 musculoskeletal tumours, 47 tumours were malignant and 23 were benign. They were located in bone (n = 46), bone and soft tissue (n = 1) and soft tissue (n = 23). Distribution of pathological diagnosis is listed in Table 2.

Table 2.

Distribution of pathological diagnosis in musculoskeletal tumours

	Benign (23)	Malignant (47)
Bone tumour (46)	Enchondroma (6)	Metastasis (25)
	Aneurysmal bone cyst (1)	Multiple myeloma/plasmacytoma (5)
	Fibrous dysplasia (1)	Lymphoma (3)
		Chondrosarcoma (2)
		Osteosarcoma (1)
		Leukaemia (1)
		Undifferentiated sarcoma (1)
Bone and soft tissue tumour (1)		Precursor B lymphoblastic leukaemia/Lymphoma (1)
Soft tissue tumour (23)	Haemangioma (4)	Leiomyosarcoma (2)
	Schwannoma (3)	Fibromyxoid sarcoma (1)
	Neurofibroma (3)	Malignant solitary fibrous tumour (1)
	Tenosynovial giant cell tumour (2)	Metastasis (1)
	Giant cell tumour of soft tissue (1)	Chronic myeloid leukaemia (1)
	Leiomyoma (1)	Squamous cell carcinoma (1)
	Intramuscular myxoma (1)	Undifferentiated sarcoma (1)

The values are the number of tumours in parentheses.

IVIM analysis for the differentiation between benign and malignant musculoskeletal tumours

D and ADC values of malignant tumours (923 ± 360, 965 ± 353 µm² s^–1, respectively) were significantly lower than those of benign tumours (1668 ± 546, 1689 ± 526 µm² s^–1, respectively) (p < 0.001, p < 0.001, respectively) (Table 3). F values of malignant tumours (9.6%) were significantly higher than those of benign tumours (7.2%) (p = 0.021), whereas D* values showed no significant difference between the malignant and benign tumours (p = 0.651). Figures 2 and 3 show exemplary images and signal curves for malignant and benign musculoskeletal tumours, respectively.

Table 3.

Comparison of ADC and IVIM parameters between benign and malignant musculoskeletal tumours

Parameter	Benign	Malignant	p-value
ADC (μm2 s–1)	1689 ± 526	965 ± 353	<0.001
D (μm2 s–1)	1668 ± 546	923 ± 360	<0.001
D* (μm2 s–1)	2629 ± 556	2552 ± 719	0.651
f (%)	7.18 ± 3.60	9.56 ± 4.16	0.021

ADC, apparent diffusion coefficient; D, pure diffusion coefficient; D*, pseudodiffusion coefficient; f, perfusion fraction; IVIM, intravoxel incoherent motion.

Figure 2.

A 28-year-old male with undifferentiated sarcoma. (a) Axial T₂ weighted image shows the hyperintense mass (arrow) with large soft tissue component in the left proximal femur. (b–e) Parametric maps (D map, D* map, f map and ADC map, respectively). Mean D, D*, f and ADC values of the lesion were 823 µm² s^–1, 2325 µm² s^–1, 5.8% and 854 µm² s^–1, respectively. ROI was traced manually within the solid portion of the tumour to encompass as much of the lesion as possible avoiding the area of haemorrhage or necrosis. (f) Signal decay curve shows rather gentle signal decay on both low and high b-values. ADC, apparent diffusion coefficient; ROI, region of interest.

Figure 3.

A 43-year-old female with schwannoma. (a) Axial fat-suppressed contrast-enhanced T₁ weighted image shows the right paravertebral mass (arrow) with intense enhancement. (b–e) Parametric maps (D map, D* map, f map and ADC map, respectively). Mean D, D*, f and ADC values of the lesion were 2100 µm² s^–1, 2495 µm² s^–1, 6.5% and 2089 µm² s^–1, respectively. ROI was traced manually within the solid portion of the tumour to encompass as much of the lesion as possible avoiding the area of haemorrhage or necrosis. (f) Signal decay curve shows rather steep signal decay on high b-values. ADC, apparent diffusion coefficient; ROI, region of interest.

In the differential diagnosis of malignant from benign tumours, D and ADC showed good diagnostic performance {AUC, 0.874 [95% CI (0.776–0.972)], 0.880 (0.781–0.979), respectively} and f showed poor diagnostic performance [AUC 0.671 (0.535–0.806)]. There was no significant difference in the diagnostic performance of D and ADC (p = 0.53) (Figure 4). Using cut-off value of 1200 µm² s^–1 in D, the sensitivity, specificity and accuracy were 91.5, 82.6 and 88.6%, respectively. Using cut-off value of 1200 µm² s^–1 in ADC, the sensitivity, specificity and accuracy were 91.5, 87.0 and 90.0%, respectively. Table 4 shows the diagnostic performance using cut-off value of 1200 µm² s^–1 in both D and ADC in each reader. Figure 5 shows the scatter plot of D values in both benign and malignant musculoskeletal tumours.

Figure 4.

The ROC curves of Dand f, and ADC for differentiating malignant from benign musculoskeletal tumours. The areas under the curves were 0.880, 0.874 and 0.671, for ADC, D and f, respectively. There was no significant difference in the diagnostic performance of D and ADC (p = 0.53). ADC, apparent diffusion coefficient; ROI, region of interest.

Table 4.

Diagnostic performance for differentiating benign and malignant musculoskeletal tumours

Parameter	Cut-off value	Reader	Sensitivity (%)	Specificity (%)	Accuracy (%)
ADC (μm2 s–1)	1200	R1	89 (42/47)	[81, 98]	83 (19/23)	[67, 98]	87 (61/70)	[79, 95]
R2	91 (43/47)	[84, 99]	87 (20/23)	[73, 100]	90 (63/70)	[83, 97]
D (μm2 s–1)	1200	R1	87 (41/47)	[78, 97]	83 (19/23)	[67, 98]	86 (60/70)	[78, 94]
R2	91 (43/47)	[84, 99]	87 (20/23)	[73, 100]	90 (63/70)	[83, 97]

ADC, apparent diffusion coefficient; D, pure diffusion coefficient.

Data are percentages, with raw data in parentheses and 95% confidence intervals in brackets.

Figure 5.

Dot plot showing the distribution of D values in benign and malignant musculoskeletal tumours. Using cut-off value of 1200 µm² s^–1 (horizontal line), the sensitivity, specificity and accuracy were 91.5, 82.6 and 88.6%, respectively.

Interobserver agreement of ADC and D were excellent: ICC = 0.972–0.989 for ADC, and ICC = 0.965–0.987 for D. Interobserver agreement of D* was poor to excellent: ICC = 0.386–0.764. Interobserver agreement of f was good to excellent: ICC = 0.588–0.841.

Subgroup analysis for the differentiation between benign and non-metastatic malignant musculoskeletal tumours

Among 47 malignant musculoskeletal tumours, 26 tumours were metastases and 21 were non-metastatic tumours. D and ADC values of non-metastatic malignant tumours (1008 ± 481, 1048 ± 472 µm² s^–1, respectively) were significantly lower than those of benign tumours (1668 ± 546, 1689 ± 526 µm² s^–1, respectively) (p < 0.001). D* and f values showed no significant difference between the nonmetastatic malignant and benign tumours (p = 0.666 and 0.329, respectively).

Analysis of false negative and false positive cases

Using cut-off value of 1200 µm² s^–1 in D, there were four false negative cases in malignant tumours: chondrosarcoma (n = 2), low-grade fibromyxoid sarcoma (n = 1) and metastasis (n = 1). Except one case of metastasis from hepatocellular carcinoma (D value, 1349 µm² s^–1), three false negative cases had myxoid or cartilaginous components with a mean D value of 2063 µm² s^–1 (2168, 1992 and 2028 µm² s^–1, respectively). Among benign and malignant musculoskeletal tumours, the highest D values more than 1800 µm² s^–1 were obtained from myxoid tumours (neurofibroma, schwannoma, fibromyxoid sarcoma, intramuscular myxoma, mean D value of 2218 µm² s^–1), cartilaginous tumours (enchondroma, chondrosarcoma, mean D value of 2047 µm² s^–1) and aneurysmal bone cyst (D value, 2763 µm² s^–1). Among four benign tumours with mean D values below the cut-off values of 1200 µm² s^–1, three lesions were giant cell tumours (mean D value, 785 µm² s^–1) and the other was fibrous dysplasia (1176 µm² s^–1).

DISCUSSION

Our study showed that IVIM-derived D and ADC were useful for differentiating malignant from benign musculoskeletal tumours at 3 T IVIM DW MRI. D and ADC showed the highest diagnostic performance in the differentiation of malignant from benign tumours, and no significant differences in diagnostic performance. There have been reports^{6, 23} that the ADC values of benign and malignant soft tissue tumours overlap and are therefore not useful to differentiate them, and this overlap is caused not only by contamination from vascular component but also by extracellular matrix component. On the other hand, other studies^{2, 5, 24} reported that DW imaging can differentiate malignant from benign musculoskeletal tumours, because the malignant tumours tend to have more impeded water diffusivity due to more cellularity than benign tumours. In addition, the extracellular matrix component affects D value, which reflects tissue diffusivity, and therefore there might be overlap in D values between malignant and benign musculoskeletal tumours.

Rijswijk et al⁵ investigated the soft tissue tumours (10 malignant and 12 benign soft tissue masses) using early IVIM DW imaging with five b-values (0, 176, 351, 526 and 701 s mm^–2) at 1.5 T. They reported that the D values of malignant tumours (1080 ± 230 µm² s^–1) were significantly lower than those of benign tumour (1710 ± 260 µm² s^–1), whereas ADC values between these groups were not significantly different (1300 ± 220 µm² s^–1 vs 1780 ± 240 µm² s^–1). They suggested that the contribution of perfusion to ADC values could be the reason for the difference. They presented the overlap of D values in the cases of myxoid malignant tumours or aggressive fibromatosis. However, in our study, ADC values showed a significant difference between malignant and benign musculoskeletal tumours. We assume that early IVIM DW imaging with five b-values and fewer b-values within the perfusion-sensitive range might be one of the reasons for the difference. However, in the scope of this study, it was not investigated whether the ADC computed from a reduced number of b-values would also result in similar diagnostic performance as the ADC based on the IVIM DW imaging scan protocol, which, therefore, remains to be examined in future works. It could be related to the small patient number with malignant soft tissue tumours including three myxoid malignant tumours (two myxoid liposarcomas and one low grade myxofibrosarcoma) out of 10 malignant soft tissue tumours in their study.⁵

Similar to our false negative cases for the malignant tumours, the previous studies^{2, 23, 24} reported that diffusion coefficients of myxoid tumours were higher than those of benign tumours because of high mucin and low collagen contents of the tumour. In addition, cartilaginous tumours have high true diffusion coefficient values due to high chondroid matrix content of the tumour in both malignant and benign tumours.² In our study, the false positive cases for the malignant tumours were giant cell tumours and fibrous dysplasia. Giant cell tumours contain histiocytes, multinucleated giant cells, hemosiderin granules and collagenous strands, which result in reducing the extracellular space with a concomitant decrease in true diffusion coefficient values and ADC.^{2, 25, 26} Fibrous dysplasia has varying T₂ signal intensities depending on the amount of bony trabeculae, cystic components, haemorrhage, cellularity and collagen.^{27, 28} Like the previous study by Hayashida et al,²⁸ in our study, fibrous dysplasia showed a cystic portion as well as T₂ hypointense solid portion, which seems to attribute to the impeded water diffusivity. Two readers of our study showed a discrepancy in measurement of D value according to ROI location with and without inclusion of cystic and fibrous tissues portion on standard MRI (1284 vs 1068 µm² s^–1). Thus, ROI should be located in the corresponding portion with high signal on high b-value images and the solid area on standard MRI. However, these false positive and false negative cases could be easily diagnosed based on standard MRI. Since IVIM DW images are interpreted combined with standard MRI in practice, the overlaps could be overcome. Aneurysmal bone cyst was the benign tumour that had the highest ADC and D values in our study like the previous study.^{28, 29}

Our study showed that D was a useful parameter in differentiating malignant from benign musculoskeletal tumours. Impeded water diffusivity due to high cellularity of malignant tumours seemed to be well-reflected except the false positive and false negative cases. However, f and D* were not significant diagnostic values in differentiation. In the previous study by Kang et al,¹⁴ f and D* were more valuable parameters in the differentiation of pancreatic adenocarcinomas from neuroendocrine tumours than were ADC and D. In the previous study by Yoon et al,³⁰ D* showed better diagnostic performance than ADC for differentiating significant hepatic fibrosis from normal or early hepatic fibrosis. The reason that f and D* were not useful parameters in our study may be heterogeneity of the benign and malignant musculoskeletal tumours because our study group could include not only hypervascular malignant and hypovascular benign tumours but also hypovascular malignant and hypervascular benign tumours. There is a need for further studies on availability of distinguishing disease from its mimickers in the cases of challenging differential diagnosis or prediction of histological grading.

Our study has several limitations. First, because this was a retrospective study, there might have been selection bias. We recruited consecutive patients who satisfied the inclusion criteria to avoid selection bias. Second, the study population was small, particularly in benign musculoskeletal tumours. This was related that our orthopaedic surgeons favour the conservative treatment for the definite benign tumours based on MRI. Third, pathological confirmation was not available in one case of fibrous dysplasia. However, we believe that clinical and imaging follow-up of fibrous dysplasia would be sufficient to conclude the benignity of the tumour. Fourth, the ROIs might include tiny mineralization or calcifications that affect ADC or IVIM-related parameters.

In conclusion, IVIM-derived D and ADC may be more accurate and reliable for the differentiation of malignant from benign musculoskeletal tumours than f and D* at 3 T IVIM DW imaging.